Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
Description
When the MonitorActivity is set up in cluster environment, with using "cluster scope" and "report on all node" behaviour, it might send out invalid activity restored flow file after reconnecting to the cluster. This might happen when the preceding processor is running only on the primary node and the disconnecting node is not the primary. In this case the preceding processor will not being triggered in the disconnected node, which if the disconnection takes longer will eventually will take enough time to reach the threshold. At this point, the internal state of the affected node's MonitorActivity will change: the inactive attribute will be set into false. With the current changes from NIFI-7959 the reporting will be prevented until the node is disconnected, but after reconnect, at the first trigger (considering there was activity in the healthy part of the cluster), as it will reach the shared state from the state map, it will change back the internal state into active and will send out a (false) report. As at the given point of time, it's part of the cluster, nothing prevents it from doing this.
In order to avoid this misbehaviour, I propose to (similar to the reporting in the mentioned ticket) prevent updating the "inactive" attribute when the node is in cluster scope setup but disconnected.